Europe is broadening its network of publicly financed AI supercomputers with a €387.8 million ($449.4 million) system from the French state-owned company Bull, in an effort to reduce the computing gap with the US and China.
Bull announced to that this is its biggest contract ever; occurring just months after the French government separated the business from the indebted Atos for as much as €404 million.
The European Union's supercomputing organization EuroHPC chose Bull to create LUMI-AI, the sixth system commissioned in its AI Factories initiative. The system is a new supercomputer that will be set up next to the current LUMI machine at Finland's supercomputing center. It is anticipated to go live in the latter part of 2027.
Funding is divided equally between EuroHPC and the LUMI AI Factory consortium, which includes Finland, the Czech Republic, Denmark, Estonia, Norway, and Poland.
Also Read: Sayari, DataExpert Join Forces to Bring Sovereign AI to Europe
The order contributes to a network that EuroHPC has developed over many years. The collaborative effort, supported by a budget of €8.2 billion for 2021-2027, has established a network of 19 AI factory centers, centered around 12 supercomputers such as Germany's Jupiter, and is now working to secure up to seven more locations.
Also Read: Google Changes EU Spam Policy to Address Antitrust Concerns
Bull will develop the infrastructure to enable large-scale AI training and inference, as well as intensive scientific simulations. AMD chips will provide computing power, competing primarily with Nvidia, while IBM will handle storage and networking will be managed by Nokia.
Also Read: NASA Launches First Telescope Named After a Female Scientist
The AI potential of European supercomputers is currently being evaluated. OVHcloud, the largest cloud provider on the continent that it had finished initial training of a foundation model on Jupiter, while Domyn, the Italian startup at the forefront of the Commission-supported EUROPA consortium, has obtained access to EuroHPC systems to train its own AI model aiming for 400 billion parameters.